is birth weight associated with childhood lymphoma? a meta-analysis

Is birth weight associated with childhood lymphoma?A meta-analysis

C. Papadopoulou, C.N. Antonopoulos, T.N. Sergentanis, P. Panagopoulou, M. Belechri and E.T. Petridou

Department of Hygiene, Epidemiology and Medical Statistics, Athens University Medical School, Athens, Greece

Several risk factors have been identified for childhood lymphomas. The purpose of this meta-analysis was to synthesize

current evidence regarding the association between birth weight with primarily the risk for non-Hodgkin lymphoma (NHL),

given its similarity to acute lymphoblastic leukemia, Hodgkin lymphoma (HL) and any category of lymphoma. Two cohort

(278,751 children) and seven case–control studies (2,660 cases and 69,274 controls) were included. Effects estimates

regarding NHL, HL and any lymphoma were appropriately pooled using fixed or random effects model in two separate

analyses: specifically, high was compared to normal or any birth weight. Similarly, low was compared to normal or any birth

weight. No statistically significant association was found between high birth weight, as compared to normal birth weight, and

risk for NHL plus Burkitt lymphoma (OR 5 1.17, 95% CI 5 0.76–1.80, random effects), HL (OR 5 0.94, 95% CI 5 0.64–1.38,

fixed effects) or any plus Burkitt lymphoma (OR 5 1.09, 95% CI 5 0.76–1.56, fixed effects). A null association emerged when

low was compared with normal birth weight for NHL plus Burkitt lymphoma (OR 5 1.07, 95% CI 5 0.71–1.62, random

effects), HL (OR 5 0.94, 95% CI 5 0.54–1.65, fixed effects) or any plus Burkitt lymphoma (OR 5 1.02, 95% CI 5 0.79–1.33,

fixed effects). Accordingly, no association was found when high or low birth weight was compared to any birth weight.

Although current evidence suggests no association, birth weight might be a too crude indicator to reveal a genuine

association of fetal growth with specific lymphoma categories; hence, there is an emerging need for use of more elaborate

proxies, at least those accounting for gestational week.

Hematological malignancies are the most common form ofchildhood cancer, accounting for almost 50% of all cancerdiagnoses.1 Genetic aberrations promoting proliferation, dif-ferentiation and apoptosis,2 Epstein-Barr virus infection,3–5

socioeconomic status,6 environmental stimuli,7,8 genetic pre-disposition9 and immunodeficiency disorders10,11 have beenrecognized as potential risk factors. Interestingly, our recentmeta-analysis has also pointed to maternal smoking as ameaningful risk factor for childhood lymphoma.12 To this

end, special interest has been devoted towards the identifica-tion of readily available markers of prenatal, intrauterineexposures that can identify infants at risk of childhoodmalignancies.13,14

Birth weight has been used as an easily applicable, directmeasure of intrauterine growth and has been linked toseveral childhood hematological malignancies.1,15–17 In partic-ular, the association of birth weight with childhood leukemiarisk has been extensively investigated during the last twodecades. In a recent meta-analysis, comprising 32 studies and16,501 cases of all leukemia types, a positive associationbetween high birth weight and acute lymphoblastic leukemia(ALL) has been reported as well as between low birth weightand acute myeloid leukemia (AML).18 Biologically plausiblemechanisms may include growth-promoting effects of insu-lin-like growth factors (IGFs) and the failure of apoptosis ofpreleukemic cells.19

Epidemiological studies focusing on the association ofbirth weight with childhood lymphomas have so far yieldedconflicting results.1,8,20–28 Small study sample size might bea reason for this discrepancy or the results might vary bycategory of lymphoma.

The aim of this meta-analysis was to synthesize currentevidence derived from case–control and cohort studiesregarding the association of birth weight with the risk ofchildhood lymphoma overall as well as by major category,i.e., primarily with non-Hodgkin lymphoma (NHL), whichshare similar characteristics with ALL, and Hodgkin lym-phoma (HL).

Key words: birth weight, childhood, lymphoma, malignancy, meta-

analysis

Abbreviations: ALL: acute lymphoblastic leukemia; AML: acute

myeloid leukemia; CI: confidence interval; HL: Hodgkin lymphoma;

IGF: insulin-like growth factor; IRR: incidence rate ratio;

NARECHEM: Nationwide Registry of Childhood Hematological

Malignancies; NHL: non-Hodgkin lymphoma; OR: odds ratio;

PRISMA: preferred reporting items for systematic reviews and meta-

analyses; UKCCS: United Kingdom Childhood Cancer Study

Grant sponsor: National and Kapodistrian University of Athens

DOI: 10.1002/ijc.26001

History: Received 28 Nov 2010; Accepted 1 Feb 2011; Online 23

Feb 2011

Correspondence to: E.T. Petridou, Department of Hygiene,

Epidemiology and Medical Statistics, Athens University Medical

School, 75 M. Asias Str. Goudi, Athens 11527, Greece,

Tel.: þ30-210-7462187, Fax: þ30-210-7462105, E-mail: epetrid@

med.uoa.gr

Epidemiology

Int. J. Cancer: 130, 179–189 (2012) VC 2011 UICC

International Journal of Cancer

IJC

Material and MethodsData collection

The present meta-analysis was conducted in accordance withthe preferred reporting items for systematic reviews andmeta-analyses (PRISMA) guidelines.29 Following a combinedcomputerized and manual systematic database search of med-ical literature, the respective publications were retrieved fromelectronic search engines (Medline, Embase, Scopus, GoogleScholar, Ovid and the Cochrane Library). Reference lists werethereafter systematically searched for relevant articles.

Types of studies, search terms, eligibility

and exclusion criteria

Eligible studies included cohort and case–control studiesexamining the association between childhood lymphoma andbirth weight. MeSH terminology used for search purposeswere [‘‘childhood’’ (all fields) or ‘‘child’’ (all fields)] and[‘‘lymphoma’’ (all fields) or ‘‘cancer’’ (all fields) or ‘‘malig-nancy’’ (all fields)] and [‘‘birth’’ (all fields) or ‘‘perinatal’’ (allfields) or ‘‘weight’’ (all fields)]. All scientific papers publishedup to November 20, 2010 were examined and no restrictionof publication language was applied.

Excluded were studies that did not refer to childhood lym-phomas or studies that did not include exploration of theassociation between childhood lymphoma and birth weight.When multiple publications on the same study populationwere identified or study populations overlapped, only thestudy of larger size was included, unless the reported out-comes were mutually exclusive. Three reviewers (C.N.A.,T.N.S. and C.D.P.) independently extracted and analyzeddata; final decision was reached by consensus. Given thatadditional data were available from the Greek NationwideRegistry of Childhood Hematological Malignancies (NARE-CHEM) database, an ad hoc analysis was undertaken com-prising incident cases of childhood lymphoma diagnoseduntil December 2008 along with their age and gendermatched controls. The corresponding results were includedin the meta-analysis; information on the recruitment processof cases and controls as well as on the collection of informa-tion by NARECHEM have been described elsewhere.8 Fur-thermore, additional, unpublished data regarding USA onnormal and high birth weight have contributed22 and werealso taken into account in the current meta-analysis.

Data extraction

Data extracted from eligible articles included first author’sname, study year, region of origin, study design, follow-upperiod, age of participants, matching factors for controls andfactors taken into account during adjustment at the multivar-iate analysis, source of information for birth weight, defini-tion and categorization of birth weight, as well as category oflymphoma and comparison group with corresponding num-ber of cases and controls (NHL, HL and any). The corre-sponding authors were contacted if the required data were

not readily available in the published article and two out ofthem responded positively.22,30

Statistical analyses

Data synthesis and treatment effects. To assess the possibilityof an underlying U-shaped curve18 describing the associationbetween birth weight and risk for childhood lymphoma twodistinct sets of analyses were performed. First, low birthweight (defined as <2.5 kg) was compared to (i) normalbirth weight (2.5–4.0 kg or 2.5–3.5 kg depending on thestudy) and (ii) any birth weight (>2.5 kg). Similarly, highbirth weight (defined as �4 kg or �3.5 kg depending on thestudy) was compared to (i) normal birth weight and (ii) anybirth weight (<4 kg or <3.5 kg, depending on the study).

Three separate subanalyses were performed, according to theoutcome measures under investigation. In particular, the associ-ation between birth weight and risk for (i) NHL, (ii) HL and(iii) any category of lymphoma was tested. Concerning the sub-analyses on NHL and HL, data synthesis was rather straightfor-ward as it was based on the relevant data from the individualstudies. Regarding any category of lymphoma, however, weopted to maximize the number of individuals included, under-taking the following scenarios: (i) when the study reported onlyany category of lymphoma data, it was evidently included onlyat the ‘‘any category’’ analysis, (ii) when the study reporteddata as any category, NHL and HL, the any category analysisevidently encompassed only the any category data.

Adjusted odds ratios (ORs) and incidence rate ratios(IRRs) with corresponding 95% confidence intervals (CIs)were extracted as appropriate from each study for anycategory of lymphoma, NHL and HL. When the above infor-mation was not available, crude ORs and 95% CIs werecalculated on the basis of the respective 2 � 2 tables.

The fixed effects model or the random effects model wasused for pooling nonheterogeneous or heterogeneous data, asappropriate. An effect estimate >1 denoted increased risk forchildhood lymphoma. The Z test was applied for the overalleffect and the statistical significance level was set at p < 0.05.Data were graphically presented as forest plots.

Heterogeneity, publication bias and sensitivity analy-

sis. Heterogeneity among studies was estimated by chi-square test and Cochran Q score (reported as I2) with corre-sponding p-values and the level of significance was set at p ¼0.10. Publication bias was assessed by Egger’s regression(Egger’s intercept),31 Kendall’s rank correlation coefficient(Kendall’s tau),31 Harbord test for small-study effects appliedonly to binary outcome data32 and visual inspection of thefunnel plots. Sensitivity analysis was performed excludingstudies or study subgroups on Burkitt lymphoma, as their ep-idemiological profiles may vary.33 Pooling of individual stud-ies through fixed or random effects models, assessment ofheterogeneity, publication bias and sensitivity analyseswere performed using the comprehensive meta-analysis v2.2(Biostat, Englewood, NJ), whereas the Harbord test was

Epidemiology

180 Is birth weight associated with childhood lymphoma?


performed using STATA v11.1 (StataCorp LP, College Sta-tion, TX).

ResultsAs shown in the flow diagram (Fig. 1), after extraction andreview of the abstracts, 15 articles1,8,16,20–28,30,34,35 were even-tually deemed relevant out of a total of 2,930 potentially

articles of interest; out of these, in the subsequent detailedevaluation, six articles were excluded.16,24–26,34,35 Specifically,the article by Schuz et al.35 had been conducted on an popu-lation overlapping with that of the larger earlier investigationconducted by the same research team21; similarly, one articleby Roman et al.16 was excluded due to its overlapping popu-lation with the larger eligible United Kingdom Childhood

Figure 1. Flow chart presenting the selection of eligible studies.

Epidemiology

Papadopoulou et al. 181


Table

1.Includedstudiesch

aracteristics

First

author

(reference)

Year

Country

Study

design

Follow-up

period

Age

range

(years)

Matching

factors

Adjusting

factors

Sourceof

inform

ation

forbirth

weight

Birth

weight(g),

definition

Typeof

lymphoma

Cases

(n)

Control

(n)

Compariso

n

groups

Effect

estim

ate,

(OR,IRR)

with95%

CI

Petridou

etal.8

Updated

Greece

Case–

control

1996–

2008

0–14

Age,sex

Maternaleducation,

smokingduring

pregnancy,birth

order,

crowdingindex,

maternal

ageatbirth

Interview

withthe

guardians

ofch

ildren

Low

<2,500,

2,500�

norm

al<

4,000,

high�

4,000

NHL

166

166

Low

vs.norm

al

OR¼

0.77(0.33–1.78)

Low

vs.any

OR¼

0.75(0.32–1.74)

Highvs.norm

al

OR¼

2.73(1.10–6.80)

Highvs.any

OR¼

2.75(1.11–6.84)

HL

111

111

Low

vs.norm

al

OR¼

0.83(0.24–2.81)

Low

vs.any

OR¼

0.82(0.24–2.76)

Highvs.norm

al

OR¼

1.05(0.37–3.04)

Highvs.any

OR¼

1.08(0.38–3.07)

Any

277

277

Low

vs.norm

al

OR¼

0.82(0.42–1.60)

Low

vs.any

OR¼

0.79(0.41–1.54)

Highvs.norm

al

OR¼

1.94(1.01–3.75)

Highvs.any

OR¼

1.96(1.02–3.78)

Rangel

etal.20

2010

Brazil

Case–

control

1984–

2008

1–17

Age,

sex

None

Medical

records,

interview

with

themother,

validated

withbirth

certificate

Low

�2,500,

2,500<

norm

al<

4,000,

high�

4,000

NHL

131

1,575

Low

vs.norm

al

OR¼

0.73(0.38–1.38)*

Low

vs.any

OR¼

0.66(0.35–1.25)*

Highvs.norm

al

OR¼

1.68(0.91–3.10)*

Highvs.any

OR¼

1.99(1.08–3.69)*

Spector

etal.22

2009

USA

Case–

control

1980–

2004

0–14

Birth

year,

sex

Birth

order,

plurality,

maternalage,

maternalrace,

state,

gestationalage

Birth

records

Low

�2,500,

2,500<

norm

al<

4,000,

high�

4,000

NHL(plusBurkitt)

824

57,610

Low

vs.norm

al

OR¼

1.25(0.80–1.94)

Low

vs.any

OR¼

1.26(0.81–1.95)

Highvs.norm

al

OR¼

0.90(0.66–1.22)

Highvs.any

OR¼

0.89(0.66–1.21)

NHL(exclBurkitt)

592

Low

vs.norm

al

OR¼

1.67(1.03–2.70)

Low

vs.any

OR¼

1.68(1.04–2.73)

Highvs.norm

al

OR¼

0.88(0.61–1.27)

Highvs.any

OR¼

0.86(0.60–1.24)

Table

1.Includedstudiesch

aracteristics(Continued)

First

author

(reference)

Year

Country

Study

design

Follow-up

period

Age

range

(years)

Matching

factors

Adjusting

factors

Sourceof

inform

ation

forbirth

weight

Birth

weight(g),

definition

Typeof

lymphoma

Cases

(n)

Control

(n)

Compariso

n

groups

Effect

estim

ate,

(OR,IRR)

with95%

CI

HL

474

Low

vs.norm

al

OR¼

1.23(0.58–2.60)

Low

vs.any

OR¼

1.24(0.58–2.62)

Highvs.norm

al

OR¼

0.92(0.56–1.52)

Highvs.any

OR¼

0.91(0.55–1.50)

Any(plusBurkitt)

1,298

Low

vs.norm

al

OR¼

1.20(0.85–1.70)

Low

vs.any

OR¼

1.22(0.86–1.71)

Highvs.norm

al

OR¼

0.88(0.69–1.12)

Highvs.any

OR¼

0.87(0.69–1.10)

Any(exclBurkitt)

1,066

Low

vs.norm

al

OR¼

1.39(0.97–2.00)

Low

vs.any

OR¼

1.41(0.98–2.03)

Highvs.norm

al

OR¼

0.87(0.66–1.13)

Highvs.any

OR¼

0.86(0.65–1.12)

Smithetal.1

2009

England,

Wales

Case–

control

1991–1996

<15

Sex,

month,

yearofbirth,

residence

Studyregion,

sexandage

atdiagnosis

Office

for

National

Statistics

Low

<2,500,

2,500�

norm

al�

4,000,

high>

4,000

NHL

220

6,337

Low

vs.norm

al

OR¼

0.77(0.41–1.44)

Low

vs.any

OR¼

0.78(0.42–1.44)*

Highvs.norm

al

OR¼

0.64(0.38–1.06)

Highvs.any

OR¼

0.71(0.43–1.17)*

HL

84

Low

vs.norm

al

OR¼

0.56(0.18–1.81)

Low

vs.any

OR¼

0.55(0.17–1.74)*

Highvs.norm

al

OR¼

0.94(0.46–1.91)

Highvs.any

OR¼

1.02(0.51–2.04)*

Any

315

Low

vs.norm

al

OR¼

0.75(0.44–1.28)

Low

vs.any

OR¼

0.74(0.44–1.25)*

Highvs.norm

al

OR¼

0.84(0.57–1.24)

Highvs.any

OR¼

0.93(0.63–1.35)*

Johnso

netal.30

2007

USA

Cohort

1959–1966

0–8

None

None

Collaborative

Perinatal

Project

Low

�2,500,

2,500<

norm

al�

4,000,

high>

4,000

Any

549,503

Low

vs.norm

al

OR¼

0.74(0.04–13.35)*

Low

vs.any

OR¼

0.78(0.04–14.16)*

Highvs.norm

al

OR¼

1.50(0.08–27.11)*

Highvs.any

OR¼

1.68(0.09–30.45)*

Leeetal.28

2004

Singapore

Cohort

1992–1998

0–5

None

Gender,

gestational

age,

birth

order,

maternalage

Singapore

National

Registryof

Birthsand

Deaths

Low

<2,500,

2,500�

norm

al�

3,500,

high>

3,500

NHL(Burkittonly)

NA

229,248

Highvs.norm

al

IRR¼

10.5

(1.1–102.6)

Table

1.Includedstudiesch

aracteristics(Continued)

First

author

(reference)

Year

Country

Study

design

Follow-up

period

Age

range

(years)

Matching

factors

Adjusting

factors

Sourceof

inform

ation

forbirth

weight

Birth

weight(g),

definition

Typeof

lymphoma

Cases

(n)

Control

(n)

Compariso

n

groups

Effect

estim

ate,

(OR,IRR)

with95%

CI

McKinneyetal.27

1999

Scotland

Case–

control

1991–1994

0–14

Age,sex,

residence

None

Medical

neonatal

records

recordedby

twotrained

abstractors

Low

<2,500,

2,500�

norm

al<

3,500,

high�

3,500

Any

44

82

Low

vs.norm

al

OR¼

2.03(0.42–9.75)

Low

vs.any

OR¼

1.95(0.46–8.21)*

Highvs.norm

al

OR¼

0.99(0.43–2.28)

Highvs.any

OR¼

0.88(0.42–1.84)*

Sch

uzetal.21

1999

Germ

any

Case–

control

1992–1997

<15

Age,sex,

district

Socioeconomic

status

Mailed

questionnaires

andtelephone

interview

Low

<2,500,

2,500�

norm

al�

4,000,

high>

4,000

NHL

230

2577

Low

vs.norm

al

OR¼

2.30(1.20–4.30)

Low

vs.any

OR¼

1.90(1.06–3.40)*

Highvs.norm

al

OR¼

0.90(0.50–1.40)

Highvs.any

OR¼

0.90(0.58–1.42)*

Yeazeletal.23

1997

USA,

Canada,

Australia

Case–

control

1982–1989

<18

Age,sex,

residence

Maternalage

atbirth,birth

order,

gestationalage

Mailed

questionnaires

High>

4,000

NHL

190

816

Highvs.any

OR¼

1.50(1.00–2.40)

Any�

4,000

HL

175

Highvs.any

OR¼

0.80(0.50–1.50)

Effect

estim

atesmarkedwithasterisk

(*)representcrudeoddsratios.

Abbreviations:

NHL:

non-Hodgkin

Lymphoma;HL:

Hodgkin

lymphoma;OR:oddsratio;IRR:incidence

rate

ratio;CI:confidence

interval.

Table

2.Resu

ltsofmeta-analyses:

Pooledeffect

estim

ates,

assessmentofheterogeneityandpublicationbiasforNHL,

HLandanylymphomaamongcompariso

ngroupsofbirth

weight

Publicationbias

Compariso

ngroups

Numberof

studies

Pooledeffect

estim

ate,

OR(95%

CIs),analysismethod

Heterogeneity

I2,pvalue

Egger’sregression

interceptb,pvalue

Kendal’stau,

pvalue

Low

vs.any

NHL(plusBurkitt)

5OR¼

1.03(0.70–1.51),random

effects

I2¼

51.27,p¼

0.08

b¼

�3.21,p¼

0.38

tau¼

�0.30,p¼

0.46

NHL(exclBurkitt)

5OR¼

1.09(0.70–1.70),random

effects

I2¼

62.63,p¼

0.03

b¼

�5.74,p¼

0.21

tau¼

�0.30,p¼

0.46

HL

3OR¼

0.94(0.54–1.64),fixedeffects

I2¼

0.00%,p¼

0.49

b¼

�2.72,p¼

0.32

tau¼

0.00,p¼

0.99

Anylymphoma(plusBurkitt)

5OR¼


I2¼

0.00%,p¼

0.43

b¼

�0.21,p¼

0.84

tau¼

0.00,p¼

0.99

Anylymphoma(exclBurkitt)

5OR¼


I2¼

28.09%,p¼

0.23

b¼

�0.38,p¼

0.77

tau¼

0.00,p¼

0.99

Low

vs.norm

al

NHL(plusBurkitt)

5OR¼

1.07(0.71–1.62),random

effects

I2¼

55.56,p¼

0.06

b¼

�2.02,p¼

0.61

tau¼

�0.10,p¼

0.81

NHL(exclBurkitt)

5OR¼

1.14(0.71–1.83),random

effects

I2¼

64.13%,p¼

0.03

b¼

�4.40,p¼

0.38

tau¼

�0.10,p¼

0.81

HL

3OR¼


I2¼

0.00%,p¼

0.52

b¼

�2.56,p¼

0.33

tau¼

0.00,p¼

0.99


5OR¼


I2¼

0.00%,p¼

0.51

b¼

�0.17,p¼

0.86

tau¼

0.00,p¼

0.99


5OR¼


I2¼

19.69%,p¼

0.29

b¼

�0.36,p¼

0.76

tau¼

0.00,p¼

0.99

Highvs.any

NHL(plusBurkitt)

6OR¼

1.18(0.84–1.67),random

effects

I2¼

66.35,p¼

0.01

b¼

3.42,p¼

0.15

tau¼

0.40,p¼

0.26

NHL(exclBurkitt)

6OR¼

1.18(0.83–1.69),random

effects

I2¼

66.23%,p¼

0.01

b¼

4.18,p¼

0.14

tau¼

0.40,p¼

0.26

HL

4OR¼


I2¼

0.00%,p¼

0.94

b¼

0.88,p¼

0.31

tau¼

0.17,p¼

0.73


5OR¼


I2¼

25.75%,p¼

0.25

b¼

1.05,p¼

0.33

tau¼

0.30,p¼

0.46


5OR¼


I2¼

25.60%,p¼

0.25

b¼

1.08,p¼

0.35

tau¼

0.30,p¼

0.46

Highvs.norm

al

NHL(plusBurkitt)

6OR¼

1.17(0.76–1.80),random

effects

I2¼

66.74%,p¼

0.01

b¼

2.66,p¼

0.08

tau¼

0.67,p¼

0.06

NHL(exclBurkitt)

5OR¼

1.07(0.72–1.60),random

effects

I2¼

63.22,p¼

0.03

b¼

3.96,p¼

0.16

tau¼

0.70,p¼

0.09

HL

3OR¼


I2¼

0.00%,p¼

0.97

b¼

0.44,p¼

0.19

tau¼

0.67,p¼

0.30


6OR¼


I2¼

48.51%,p¼

0.08

b¼

1.47,p¼

0.12

tau¼

0.40,p¼

0.26


5OR¼


I2¼

26.14%,p¼

0.25

b¼

0.05,p¼

0.35

tau¼

0.50,p¼

0.22

Abbreviations:

NHL:

non-Hodgkin

lymphoma;HL:

Hodgkin

lymphoma;OR:oddsratio;CI:confidence

interval.

Epidemiology



Cancer Study (UKCCS) by Smith et al.1; lastly, the earlierNARECHEM article26 was excluded due to overlap with thelatest data presented herein. Three articles were thereafterexcluded for reporting reasons, namely, the article by Milneet al.,25 because the proportion of optimal birth weight wastreated as a continuous variable; the article by McKinney etal.,24 because it provided only mean birth weight values forcases and controls; a second article by Roman et al.,34 as itdid not make the distinction between children and youngadults when reporting the results. Results from one of the eli-gible article8 were also appropriately replaced by the updatedNARECHEM data of the same investigating group. Eventu-ally, the 9 published articles corresponded to 13 individualstudies, as more than one studies (NHL, HL or any lym-phoma) have been presented per article. More specifically,among the eligible studies, data were abstracted for NHLfrom seven studies1,8,20–23,28 for HL from four studies,1,8,22,23

whereas two studies27,30 did not make the distinction betweenNHL and HL, presenting only the any category of lymphomadata.

Characteristics of eligible studies

Among the nine eligible articles, eight were published as orig-inal articles1,8,20–23,27,30 whereas one was published as a Letterto the Editor.28 Seven articles used a case–control design(2,660 cases and 69,274 controls)1,8,20–23,27 and two a cohortdesign (278,751 children).28,30 The follow-up period, regionof origin, age range, matching and adjusting factors, sourceof information for birth weight, definition of birth weightcategories and numbers of cases and controls, as well aseffect estimates are shown in Table 1.

Meta-analyses

Low birth weight and risk of childhood lymphoma. The asso-ciation between low birth weight and the risk for childhood

lymphoma was not statistically significant, as shown in theanalysis pertaining to the <2.5 kg vs. �2.5 kg comparison(low vs. any) (Table 2), which allowed inclusion of the maxi-mum number of studies.1,8,20–22,27,30 In particular, the pooledORs were as follows: at the analysis on NHL (plus Burkittlymphoma), OR ¼ 1.03, 95% CI ¼ 0.70–1.51, random effectsmodel, I2 ¼ 51.27%, p for heterogeneity ¼ 0.08; in the analy-sis on HL, OR ¼ 0.94, 95% CI ¼ 0.54–1.64, fixed effectsmodel, I2 ¼ 0.0%, p for heterogeneity ¼ 0.49 and in the‘‘any’’ plus Burkitt lymphoma analysis, OR ¼ 1.03, 95% CI ¼0.79–1.33, fixed effects model, I2 ¼ 0.0%, p for heterogeneity¼ 0.43.

The analysis on the subset of studies1,8,20,21,27,30 allowingthe direct comparison between low and normal birth weight(low vs. normal) (Table 2) confirmed the findings presentedabove. In particular, the pooled ORs were as follows: in theanalysis on NHL (plus Burkitt lymphoma), OR ¼ 1.07, 95%CI ¼ 0.71–1.62, random effects model, I2 ¼ 55.56%, p forheterogeneity ¼ 0.06, Figure 2; in the analysis on HL, OR ¼0.94, 95% CI ¼ 0.54–1.65, fixed effects model, I2 ¼ 0.0%, pfor heterogeneity ¼ 0.52, Figure 3 and in the ‘‘any’’ plus Bur-kitt lymphoma analysis, OR ¼ 1.02, 95% CI ¼ 0.79–1.33,fixed effects model, I2 ¼ 0.0%, p for heterogeneity ¼ 0.51.

The results remained practically unchanged in the sensi-tivity analysis of both low vs. any and low vs. normal birthweight excluding the study arm encompassing Burkitt lym-phoma.22 Publication bias was not significant in the NHL,HL or ‘‘any lymphoma’’ analysis.

High birth weight and risk of childhood lymphoma. Theassociation between high birth weight and the risk for child-hood lymphoma was also not statistically significant, asshown in the analysis based on the �4 kg versus <4 kg com-parison (high vs. any) (Table 2), which allowed inclusion of

Figure 2. Forest plot presenting the meta-analysis based on ORs for the association between low vs. normal birth weight and risk for

childhood NHL. ORs in the individual studies are presented as squares with 95% CIs presented as extending lines. The pooled OR with its

95% CI is depicted as a diamond.

Epidemiology



the maximum number of studies.1,8,20–23,27,30 In particular,the pooled ORs were as follows: in the analysis on NHL(plus Burkitt lymphoma), OR ¼ 1.18, 95% CI ¼ 0.84–1.67,random effects model, I2 ¼ 66.35%, p for heterogeneity ¼0.01; in the analysis on HL, OR ¼ 0.92, 95% CI ¼ 0.66–1.24,fixed effects model, I2 ¼ 0.0%, p for heterogeneity ¼ 0.94and in the ‘‘any’’ plus Burkitt lymphoma analysis, OR ¼0.95, 95% CI ¼ 0.79–1.14, fixed effects model, I2 ¼ 25.75%, pfor heterogeneity ¼ 0.25.

The analysis on the subset of studies1,8,20–23,27 allowingthe direct comparison between high and normal birth weight(high vs. normal) (Table 2) confirmed the findings presentedabove. In particular, the pooled ORs were as follows: in the

analysis on NHL (plus Burkitt lymphoma), OR ¼ 1.17, 95%CI ¼ 0.76–1.80, random effects model, I2 ¼ 66.74%, pfor heterogeneity ¼ 0.01, Figure 4; in the analysis on HL,OR ¼ 0.94, 95% CI ¼ 0.64–1.38, fixed effects model, I2 ¼0.0%, p for heterogeneity ¼ 0.97 and in the ‘‘any’’ plusBurkitt lymphoma analysis, OR ¼ 1.09, 95% CI ¼ 0.76–1.56, fixed effects model, I2 ¼ 48.51%, p for heterogeneity ¼0.08.

The results remained practically unchanged in the sensi-tivity analysis of both high vs. any and high vs. normal birthweight excluding the study arm encompassing Burkitt lym-phoma.28 Once more, publication bias was not significant inthe NHL, HL or any lymphoma analysis.

Figure 3. Forest plot presenting the meta-analysis based on ORs for the association between low vs. normal birth weight and risk for

childhood HL. ORs in the individual studies are presented as squares with 95% CIs presented as extending lines. The pooled OR with its


Figure 4. Forest plot presenting the meta-analysis based on ORs for the association between high vs. normal birth weight and risk for

childhood NHL. ORs in the individual studies are presented as squares with 95% CIs presented as extending lines. The pooled OR with its


Epidemiology



DiscussionThe present meta-analysis demonstrated no evidence of astatistically significantly increased risk for NHL among chil-dren born with high vs. normal birth weight (OR ¼ 1.17),despite our prior hypothesis. The associations of NHL werepractically null regarding children with low vs. normal birthweight (OR ¼ 1.07) and likewise of HL for either high orlow birth weights (OR ¼ 0.94, for both high vs. normal andlow vs. normal birth weight comparisons) and subsequentlyany lymphoma (OR ¼ 1.09 and OR ¼ 1.02, for high and lowvs. normal birth weight, respectively). Among the articlesincluded in the meta-analysis, three studies reported statisti-cally significant positive associations between NHL and highbirth weight,20,23,28 while one supported a link between NHLand low birth weight.21 This was in stark contrast with theobserved associations of birth weight with HL, as no statisti-cal association had emerged with this lymphoma category inany individual study.1,8,23

Inconsistencies were also noted among studies reportingon birth weight in relation to childhood leukemia until arecent meta-analysis demonstrated a positive and statisti-cally significant association.18 Childhood lymphoma andleukemia share common morphological and immunopheno-typical characteristics. In particular, because mature cell leu-kemia is very rare in children, most similarities have beenreported between precursor-cell ALL and precursor-cell(lymphoblastic) NHL, which represents up to 30% of allchildhood lymphomas. The lymphoblasts in T-ALL/lympho-blastic lymphoma (LBL) are similar to precursor B lympho-blasts, of medium size with a high nuclear cytoplasmic ra-tio, TdT positive and variably express of CD1a, CD2, CD3,CD4, CD5, CD7 and CD8. The clinical presentation andcytogenetic characteristics are largely similar for T-ALL andT-LBL supporting the concept that both may represent aspectrum of one single disease. Despite commonalities inmorphological and immunophenotypical characteristics2,36

of the two disease entities, reasons for the discrepancybetween the findings of the respective meta-analyses mayencompass a physiological component, which remains elu-sive for the time being. Alternatively, it might be that birthweight is inherently too crude a proxy of intrauterinegrowth to also reveal an association with NHL, whichmight be of less size compared to that with ALL. Addition-ally, leukemia is by far more prevalent than childhood lym-phoma and the studies to be meta-analyzed were lessnumerous. Therefore, there was less statistical power in thepresent meta-analysis compared to the respective for leuke-mia, the latter including nearly an eightfold larger poolthan ours. Issues related to the validity of lymphoma cate-gorization implemented in this meta-analysis are also aconcern, as the well-established categories of NHL andHL37 include a spectrum of diverse histological subtypeswith variable for example epidemiological profiles in chil-dren compared to adults. Worthy of note, due to lack of

data in the constituent articles, we were unable to proceedwith subtype-specific summary estimates.

Birth weight is actually a multicomponent mixture model,depending on the gestational age and fetal growth velocity;hence, examination of each constituent is necessary to yieldmore elaborate information regarding its association withlymphomas and NHL, in particular. To this direction, Milneet al.25 used three neonatal and anthropometrical indices asproxy measures of intrauterine growth; proportion of optimalbirth weight, birth length and weight for length. The resultswere promising, suggesting that accelerated intrauterinegrowth may be more important than birth weight per se.Interestingly, the risk of HL was found higher—among girlsonly—primarily with greater skeletal growth but also withsomewhat more soft tissue at birth than expected, while therisk of HL was associated with higher than expected birthweight, birth length, and birth weight for length, particularlyin boys.

This meta-analysis succinctly combined several categoricalbirth weight variables such as 3.5 kg27,28 or 4.0 kg1,8,20,21,23,30

for all lymphomas categories (NHL, HL and any). Despitethe apparently heterogeneous categorization in the individualstudies, statistically significant heterogeneity did not seem tomaterially impact on the validity of the analysis, which wasobserved only in the NHL subanalysis and thereafter appro-priately treated. Noticeably, not all studies have been adjustedfor a potential confounders. Five studies1,8,21,23,28 adjusted forsocioeconomic status, birth order and maternal age, but onlythree studies22,23,28 gestational age, another two20,21 formaternal smoking, an established risk factor for low birthweight and none for genetic confounders. Publication biasseemed minimal, whereas it is unlikely that recall bias playeda significant role, as the individual studies included werebased on registries or medical records, except for three whichwere based on interviews.8,21,23 On the contrary, the smallnumber of studies (three to five) in each lymphoma subcate-gory, is among the inherent limitations of the meta-analysis,as it depends on individual studies currently available.

In conclusion, this meta-analysis points to a non statisti-cally significant positive association of high birth weightwith childhood NHL and null associations in the otherbirth weight groups and disease categories examined. Giventhe rarity of the disease, the paucity of publications andthe inherent limitations of birth weight in assessing fetalgrowth and velocity, there is an emerging need for use ofmore elaborate proxies, at least those accounting for gesta-tional week.

AcknowledgementsThe authors acknowledge the contributions of Drs. Logan G. Spector, SusanA. Carozza, Colleen McLaughlin, Beth A. Mueller and Peggy Reynolds forcontributing crude and adjusted odds ratios included in the meta-analysis,as derived from their data and Dr. Kara J. Johnson for providing additionaldata.

Epidemiology



References

1. Smith A, Lightfoot T, Simpson J, RomanE. Birth weight, sex and childhood cancer:a report from the United KingdomChildhood Cancer Study. Cancer Epidemiol2009;33:363–7.

2. Link MP, Weinstein HJ. Malignant Non-Hodgkin lymphomas in children. In: PizzoPA, Poplack DG. Principles and practice ofpediatric oncology, 5 ed. Philadelphia:Lippincott, Williams & Wilkins, 2006.

3. Ambinder RF, Lemas MV, Moore S, Yang J,Fabian D, Krone C. Epstein-Barr virus andlymphoma. Cancer Treat Res. 1999;99:27–45.

4. Hjalgrim H, Askling J, Sorensen P, MadsenM, Rosdahl N, Storm HH, Hamilton-DutoitS, Eriksen LS, Frisch M, Ekbom A, MelbyeM. Risk of Hodgkin’s disease and othercancers after infectious mononucleosis. JNatl Cancer Inst 2000;92:1522–8.

5. Niedobitek G, Young LS, Herbst H.Epstein-Barr virus infection and thepathogenesis of malignant lymphomas.Cancer Surv 1997;30:143–62.

6. Stiller CA. What causes Hodgkin’s diseasein children? Eur J Cancer 1998;34:523–8.

7. John EM, Savitz DA, Sandler DP. Prenatalexposure to parents’ smoking andchildhood cancer. Am J Epidemiol 1991;133:123–32.

8. Petridou ET, Dikalioti SK, Skalkidou A,Andrie E, Dessypris N, Trichopoulos D.Sun exposure, birth weight, and childhoodlymphomas: a case control study in Greece.Cancer Causes Control 2007;18:1031–7.

9. Mack TM, Cozen W, Shibata DK, WeissLM, Nathwani BN, Hernandez AM, TaylorCR, Hamilton AS, Deapen DM, RappaportEB. Concordance for Hodgkin’s disease inidentical twins suggesting geneticsusceptibility to the young-adult form ofthe disease. N Engl J Med 1995;332:413–18.

10. Kersey JH, Shapiro RS, Filipovich AH.Relationship of immunodeficiency tolymphoid malignancy. Pediatr Infect Dis J1988;7:S10–2.

11. McClain KL, Joshi VV, Murphy SB.Cancers in children with HIV infection.Hematol Oncol Clin North Am 1996;10:1189–201.

12. Antonopoulos CN, Sergentanis TN,Papadopoulou C, Andrie E, Dessypris N,Panagopoulou P, Polychronopoulou S,Pourtsidis A, Athanasiadou-PiperopoulouF, Kalmanti M, Sidi V, Moschovi M, et al.Maternal smoking during pregnancy andchildhood lymphoma: a meta-analysis. Int JCancer 2011.

13. Brooks AA, Johnson MR, Steer PJ, PawsonME, Abdalla HI. Birth weight: nature ornurture? Early Hum Dev 1995;42:29–35.

14. Petridou E, Trichopoulos D, Revinthi K,Tong D, Papathoma E. Modulation ofbirthweight through gestational age and

fetal growth. Child Care Health Dev 1996;22:37–53.

15. Okcu MF, Goodman KJ, Carozza SE,Weiss NS, Burau KD, Bleyer WA, CooperSP. Birth weight, ethnicity, and occurrenceof cancer in children: a population-based,incident case-control study in the State ofTexas, USA. Cancer Causes Control 2002;13:595–602.

16. Roman E, Simpson J, Ansell P, LightfootT, Mitchell C, Eden TO. Perinatal andreproductive factors: a report onhaematological malignancies from theUKCCS. Eur J Cancer 2005;41:749–59.

17. Daling JR, Starzyk P, Olshan AF, WeissNS. Birth weight and the incidence ofchildhood cancer. J Natl Cancer Inst 1984;72:1039–41.

18. Caughey RW, Michels KB. Birth weightand childhood leukemia: a meta-analysisand review of the current evidence. Int JCancer 2009;124:2658–70.

19. Greaves M. Molecular genetics, naturalhistory and the demise of childhoodleukaemia. Eur J Cancer 1999;35:1941–53.

20. Rangel M, Cypriano M, de Martino LeeML, Luisi FA, Petrilli AS, Strufaldi MW,Franco Mdo C. Leukemia, non-Hodgkin’slymphoma, and Wilms tumor inchildhood: the role of birth weight. Eur JPediatr 2010;169:875–81.

21. Schuz J, Kaatsch P, Kaletsch U, Meinert R,Michaelis J. Association of childhoodcancer with factors related to pregnancyand birth. Int J Epidemiol 1999;28:631–9.

22. Spector LG, Puumala SE, Carozza SE,Chow EJ, Fox EE, Horel S, Johnson KJ,McLaughlin CC, Reynolds P, Behren JV,Mueller BA. Cancer risk among childrenwith very low birth weights. Pediatrics2009;124:96–104.

23. Yeazel MW, Ross JA, Buckley JD, WoodsWG, Ruccione K, Robison LL. High birthweight and risk of specific childhoodcancers: a report from the Children’sCancer Group. J Pediatr 1997;131:671–7.

24. McKinney PA, Cartwright RA, Saiu JM,Mann JR, Stiller CA, Draper GJ, HartleyAL, Hopton PA, Birch JM, Waterhouse JA,Johnston HE. The inter-regionalepidemiological study of childhood cancer(IRESCC): a case control study ofaetiological factors in leukaemia andlymphoma. Arch Dis Child 1987;62:279–87.

25. Milne E, Laurvick CL, Blair E, de Klerk N,Charles AK, Bower C. Fetal growth andthe risk of childhood CNS tumors andlymphomas in Western Australia. Int JCancer 2008;123:436–43.

26. Petridou E, Andrie E, Dessypris N,Dikalioti SK, Trichopoulos D. Incidenceand characteristics of childhood Hodgkin’s

lymphoma in Greece: a nationwide study(Greece). Cancer Causes Control 2006;17:209–15.

27. McKinney PA, Juszczak E, Findlay E,Smith K, Thomson CS. Pre- andperinatal risk factors for childhoodleukaemia and other malignancies: aScottish case control study. Br J Cancer1999;80:1844–51.

28. Lee J, Chia KS, Cheung KH, Chia SE, Lee HP.Birthweight and the risk of early childhoodcancer among Chinese in Singapore. Int JCancer 2004;110:465–7.

29. Liberati A, Altman DG, Tetzlaff J, MulrowC, Gotzsche PC, Ioannidis JP, Clarke M,Devereaux PJ, Kleijnen J, Moher D. ThePRISMA statement for reporting systematicreviews and meta-analyses of studies thatevaluate health care interventions:explanation and elaboration. J ClinEpidemiol 2009;62:e1–34.

30. Johnson KJ, Spector LG, Klebanoff MA,Ross JA. Childhood cancer and birthmarksin the Collaborative Perinatal Project.Pediatrics 2007;119:e1088–93.

31. Higgins JPT, Green S. Cochrane handbook forsystematic reviews of interventions Version5.0.2 [updated September 2009] 2009.Available at: www.cochrane-handbook.org..

32. Harbord RM, Egger M, Sterne JA. Amodified test for small-study effects inmeta-analyses of controlled trials withbinary endpoints. Stat Med 2006;25:3443–57.

33. Mbulaiteye SM, Biggar RJ, Bhatia K, LinetMS, Devesa SS. Sporadic childhood Burkittlymphoma incidence in the United Statesduring 1992–2005. Pediatr Blood Cancer2009;53:366–70.

34. Roman E, Ansell P, Bull D. Leukaemiaand non-Hodgkin’s lymphoma in childrenand young adults: are prenataland neonatal factors important determinantsof disease? Br J Cancer1997;76:406–15.

35. Schuz J, Forman MR. Birthweight bygestational age and childhood cancer. CancerCauses Control 2007;18:655–63.

36. Burkhardt B, Moericke A, Klapper W,Greene F, Salzburg J, Damm-Welk C,Zimmermann M, Strauch K, Ludwig WD,Schrappe M, Reiter A. Pediatricprecursor T lymphoblasticleukemia and lymphoblastic lymphoma:differences in the commonregions with loss of heterozygosity atchromosome 6qand their prognostic impact. LeukLymphoma 2008;49:451–61.

37. Adami H, Hunter D, Trichopoulos DA.Textbook of cancer epidemiology.New York: Oxford UniversityPress, 2002.

Epidemiology



is birth weight associated with childhood lymphoma? a meta-analysis

Documents